Growing Action Spaces
Authors: Gregory Farquhar, Laura Gustafson, Zeming Lin, Shimon Whiteson, Nicolas Usunier, Gabriel Synnaeve
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show the efficacy of our approach in proof-of-concept control tasks and on challenging large-scale Star Craft micromanagement tasks with large, multi-agent action spaces. |
| Researcher Affiliation | Collaboration | 1University of Oxford 2Work done at Facebook AI Research 3Facebook AI Research. |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/ Torch Craft/Torch Craft AI/tree/gas-micro. |
| Open Datasets | No | The paper uses Star Craft micromanagement scenarios and custom environments like Acrobot and Mountain Car. While these are common benchmarks in RL, the paper does not provide specific access information (link, DOI, formal citation) to a publicly available *dataset* in the traditional sense, as data is generated dynamically through interaction with the environment rather than loaded from a static dataset. |
| Dataset Splits | No | The paper does not provide specific training/validation/test dataset splits (e.g., percentages or counts) as one would for a static dataset. It discusses evaluation during training but not explicit data partitioning for reproducibility. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running experiments, such as CPU or GPU models, or memory specifications. |
| Software Dependencies | No | The paper mentions "Torch Craft" in the code link but does not list any specific software dependencies or libraries with version numbers (e.g., Python 3.x, PyTorch 1.x). |
| Experiment Setup | Yes | We use a simple linear schedule on a mixing parameter α [0, N]. Initially α = 0 and we always choose ℓ= 0. Later, we pick ℓ= α with probability α α and ℓ= α with probability α α (e.g. if α = 1.1, we choose ℓ= 1 with 90% chance and ℓ= 2 with 10% chance). |